The Five Horsemen of Prompt Injection: A Technical Deep Dive into LLM Attack Vectors

<h3>Prevention Mechanism: Conversation Policy Enforcement with State Validation</h3> <p><strong>Strategy</strong>: Explicitly track and re-validate the conversation state and policy at each turn.</p>

python
 
1from dataclasses import dataclass
2from typing import List, Optional
3from enum import Enum
4from datetime import datetime
5
6class TurnRole(Enum):
7    USER = "user"
8    ASSISTANT = "assistant"
9    SYSTEM = "system"
10
11@dataclass
12class ConversationTurn:
13    role: TurnRole
14    content: str
15    timestamp: datetime
16    turn_number: int
17    validated_instructions: Optional[str] = None
18
19@dataclass
20class ConversationPolicy:
21    """Defines immutable rules for the conversation"""
22    system_instructions: str
23    allowed_functions: List[str]
24    max_turns: int
25
26def validate_turn(turn: ConversationTurn, policy: ConversationPolicy) -> bool:
27    """
28    Validates a turn against the conversation policy.
29    
30    Args:
31        turn (ConversationTurn): The current turn to be validated.
32        policy (ConversationPolicy): The immutable rules of the conversation.
33
34    Returns:
35        bool: True if the turn is valid according to the policy, False otherwise.
36
37[Read the full article at Towards AI - Medium](https://pub.towardsai.net/the-five-horsemen-of-prompt-injection-a-technical-deep-dive-into-llm-attack-vectors-56b7b9f6cbab?source=rss----98111c9905da---4)
38
39---
40
41**Want to create content about this topic?** [Use Nemati AI tools](https://nemati.ai) to generate articles, social posts, and more.